The Timing and Tempo of the Neolithic Expansion Across the Central

Journal of Archaeological Science: Reports 33 (2020) 102528

Contents lists available at ScienceDirect

Journal of Archaeological Science: Reports

journal homepage: www.elsevier.com/locate/jasrep

The timing and tempo of the Neolithic expansion across the Central Balkans T in the light of the new radiocarbon evidence ⁎ Marko Porčića,b, , Tamara Blagojevićb, Jugoslav Pendićb, Sofija Stefanovićb,c a Department of Archaeology, Faculty of Philosophy, University of Belgrade, Čika Ljubina 18-20, 11000 Belgrade, Serbia b BioSense Institute, University of Novi Sad, Serbia c Faculty of Philosophy, University of Belgrade, Serbia

ARTICLE INFO ABSTRACT

Keywords: The new set of radiocarbon dates was used to explore the timing and tempo of the Neolithic expansion across the Neolithic Central Balkans. Our results suggest that the first farmers arrived in this region around or few decades before Balkans 6200 cal BC. The observed spatio-temporal pattern based on the radiocarbon data suggests that the general Farming expansion direction of the expansion was along the south-north axis. The regression analysis (arrival time vs. distance from Radiocarbon the origin of expansion in northern Greece) was used to estimate the Neolithic front speed. The results of this analysis suggest that there is a moderate fit of the linear model. Most of the front speed estimates basedonthe Central Balkan data are between 1 and 2.5 km/year (depending on the data subset and the statistical technique) which is mostly above the expected range (around 1 km/year) for the standard wave of advance model and the empirically determined continental averages. We conclude that the spatio-temporal pattern of the Neolithic expansion in the Central Balkans is broadly consistent with the predictions of the wave of advance model, with the possibility of sporadic leapfrog migration events. The speed of the expansion seems to have been faster in the Central Balkans compared to the continental average.

1. Introduction Central Balkans, the Mesolithic communities lived in the microregion of the Danube Gorges from the beginning of the Holocene. Upon the ar- The Central Balkans was the bridge for the spread of farming from rival of farmers in the region, the Mesolithic and Neolithic communities its initial foothold in Greece to the mainland of Europe. The earliest came into contact which included the exchange of knowledge, animals European Neolithic is found in the Aegean on the sites such as Knossos and people (Borić, 2011). Both strontium and aDNA evidence suggest on the island of Crete, and Franchti cave on the Peloponnese, both that Neolithic women of Anatolian genetic ancestry came to live in the dated to the first quarter of the 7th millennium (Perlès et al., 2013; Mesolithic communities of the Gorges (Borić and Price, 2013; Perlès, 2001; Douka et al., 2017). In the northern Greece, the Neolithic Mathieson et al., 2018). started slightly later, around 6600–6500 cal BC (Karamitrou-Mentessidi Previous research on the timing, tempo and mode of the Neolithic et al., 2015; Reingruber et al., 2017), from where it spread further to expansion across the Central Balkans was based on a relatively low the Central Balkans. The aDNA research indicates that this was mainly a number of radiocarbon dates. The most recent comprehensive study demographic process with populations from the Anatolia and the Ae- was by Whittle et al. (2002) who concluded that the earliest farmers gean migrating north (Mathieson et al., 2018; Hofmanová et al., 2016). reached the Central Balkans around 6200 cal BC. As for the mode of In culture-historical terms, the earliest Central Balkan Neolithic is expansion, Whittle et al. (2002) rejected the Wave of Advance (WoA) represented by the Starčevo culture, a part of the wider Early Neolithic model (Ammerman and Cavalli-Sforza, 1973, 1984) and suggested that cultural Starčevo-Körös-Criş complex of the late 7th and 6th millen- the directed pioneer colonization was more probable. nium BC, with characteristic globular painted pottery, clay figurines, pit As for the tempo of the expansion, in the pioneering study on de- houses and small settlements (Garašanin, 1982; Gatsov and termining the rate of the Neolithic expansion in Europe by using re- Boyadzhiev, 2009; Luca and Suciu, 2011; Anders and Siklósi, 2012). gression analysis with radiocarbon dates and distances from the as- The presence of the Mesolithic in the Balkans seems to be limited only sumed origin of the Neolithic expansion as key variables, Ammerman to certain microregional pockets (Gurova and Bonsall, 2014). In the and Cavalli-Sforza (1971) found a good fit of the linear model (with

⁎ Corresponding author. E-mail address: [email protected] (M. Porčić). https://doi.org/10.1016/j.jasrep.2020.102528 Received 26 May 2020; Received in revised form 7 August 2020; Accepted 16 August 2020 2352-409X/ © 2020 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/). M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528 correlation coefficients ranging from 0.83 to 0.89, depending onthe limited our study to the territory of Serbia and Kosovo. First, we col- assumed origin of expansion), and estimated that the average con- lected all existing radiocarbon dates from the literature, and the most tinental rate was 1.08 km/year. Gkiasta et al. (2003) found that the recent dates from the BIRTH project for the Early Neolithic sites in overall continental average was ~1.3 km/year, also with a relatively Serbia into a single database. In the second step, we filtered out all good fit of the linear model (correlation coefficient between datesand dates with standard error greater than 100 radiocarbon years. In this distances from the origin was 0.74). Pinhasi et al. (2005) estimated that way, we assure that only dates with acceptable precision are used for the continental average rate of expansion was in the range between 0.6 the analysis. and 1.3 km/year and concluded that these values perfectly fit the To answer the first and second research questions, regarding the predictions of the WoA model both in terms of the actual estimated timing and spatial pattern of the expansion, we try to estimate which values, as well as the correlation coefficients measuring the goodness of are the earliest dates from the earliest sites. As many of the Early fit of dates vs. distance regressions (~0.8 in their case). Bocquet-Appel Neolithic Starčevo sites were founded long after the Neolithic had ex- et al. (2012) calculated the overall continental average rate to be panded into the particular region, we specified the following criteria for 1.09 km/year. Similar value was obtained in the latest study by the inclusion of sites as the earliest. As the main direction of the spread Henderson et al. (2014) who found 0.96 ± 0.04 km/year to be the of the Neolithic front was to the north, we assume that once the continental average. Estimated regional expansion rates vary from Neolithic front had passed through further to the north, the regions 0.7 km/year, for the Balkans, to 5.59 km/year for the LBK expansion behind the front are considered as Neolithic, even though individual area in the seminal study of Ammerman and Cavalli-Sforza (1971). settlements might have been founded centuries afterwards. For ex- Bocquet-Appel et al. (2012) regional estimates for the Balkans is ample, if we know that there are Neolithic sites north of the Sava and 0.788 km/year, close to the original Ammerman and Cavalli-Sforza’s Danube river lines dated to ~6000 cal BC, then we would exclude from (1971) estimate. Biagi et al. (2005) estimated 3.64 km/year for the our data a site that is situated south of this line if its start is dated to front speed across the Balkans, which is very high, comparable to ~5600 cal BC, as this date is not informative on the arrival of the Dolukhanov et al. (2005) estimated speed of the expansion of the LBK Neolithic. In order to solve this problem, we first look at the earliest Neolithic across the central Europe (> 4 km/year), and to Zilhão’s Neolithic radiocarbon dates in Hungary which is immediately to the (2003) estimate for the Mediterranean maritime route of expansion north of our study area. The earliest date in Hungary has a median of ranging from 3.8 and 5 km/year. 5999 cal BC when calibrated. Therefore, we only included sites for In summary, the results of previous studies established the general which the medians of the calibrated distributions of their respective outlines of the Neolithization in the Central Balkans, but many im- earliest radiocarbon dates are not later than 5999 cal BC. The more portant aspects of the process remain unknown mainly due to the low detailed description of the protocol for the selection of the sample is number of radiocarbon dates. For the purposes of the ERC BIRTH presented in Supplementary material 1 file. This sample consists of26 project a new set of 300 AMS radiocarbon samples was collected and radiocarbon dates from 26 sites (Fig. 1, Table 1). Fourteen of these dated (Porčić et al., 2020). This resulted in an unprecedented data set of dates are the new radiocarbon dates generated by the BIRTH project, new absolute dates for the Early Neolithic in the Central Balkans. In this and the remaining 12 are the dates from the literature. It should be study we combine this new high-resolution radiocarbon data set with emphasized that the selection criterion, as defined here, cannot guar- legacy dates of sufficient quality to answer three major questions: antee that sites which in reality were not truly the earliest (we will refer to them as false earliest) in their respective microregions will not be 1. When did the Neolithic arrive to the Central Balkans? We use the new included in the set. In practice, it is not possible to get a pure set of the radiocarbon evidence to update and revise previous knowledge truly earliest sites regardless of the selection criterion, as these sites about the arrival of farming to this region. might have never been archaeologically detected. The imposition of the 2. What was the route of the expansion? The aim is to reconstruct a threshold only reduces the probability of including the false earliest spatio-temporal pattern of the spread in terms of geographic spe- sites, therefore our selection would be our estimate about which sites cifics of the process. are the earliest in their microregions. That is why we refer to this 3. What was the speed of the expansion? sample as the estimated earliest sites/dates set. We calibrated the radiocarbon dates from this set in the OxCal The questions of speed and spatial patterns bear important im- software v4.3.2 (Bronk Ramsey, 2009). In order to explore the spatial plications for the reconstruction of the social, demographic, and eco- patterns of the spread, we divided the study area into 10x10km quad- nomic aspects of the Neolithic expansion, as they figure prominently in ratic grid. Each cell in this grid is associated with the earliest date from theories and models regarding the spread of the farming populations that cell. The cells are classified into 50 years wide temporal intervals into and across Europe (Ammerman and Cavalli-Sforza, 1971, 1973, on a calendric timescale, based on the means of the calibrated dis- 1984; Zvelebil, 2001; Hazelwood and Steele, 2004; Pinhasi et al., 2005; tributions of the earliest dates from each cell. The cells belonging to the Robb and Miracle, 2007; Fort et al., 2012; Fort, 2012; Bocquet-Appel same temporal interval are plotted in the same color of the color gra- et al., 2012; Shennan, 2018). In the standard WoA model, the speed of dient spectrum in order to facilitate the identification of patterns on the the expansion depends on the migration distance and the intrinsic po- map. pulation growth rate (Ammerman and Cavalli-Sforza: 68, 1984; As for the third research question, we estimated the average speed Hazelwood and Steele, 2004). Therefore, knowing the speed of the of the expansion of the Neolithic across the Central Balkans by using a expansion gives us insight into the range of possible combinations of regression analysis that involves the great circle distance from the migration distances and growth rate values (Ammerman and Cavalli- origin of the expansion and the time of arrival of the Neolithic front Sforza, 1984: 81). Admittedly, this is a rather limited information about (Ammerman and Cavalli-Sforza, 1971; Gkiasta et al., 2003; Pinhasi the details of the expansion mechanism, but if the estimated regional et al., 2005; Steele, 2010; Brami and Zanotti, 2015). The time variable front speed is outside the limits predicted by theory and/or empirically is measured for each site as the earliest date associated with the site. As established continental averages, this may signal that there may be for the distance, the first step was to determine the location ofthe regional specifics in the way that the farming was spreading. origin of the expansion. Until recently, it was considered that the origin of the spread of the Neolithic from Greece to the rest of the Balkans was 2. Materials and methods the region of Thessaly where the Neolithic settlements appeared ~6500 cal BC (Reingruber et al., 2017), but in the light of new radio- As the core of the Central Balkan region almost completely overlaps carbon evidence from the sites Paliambela and Mavropigi, both in with the boundaries of the Republic of Serbia, for practical purposes we Greek Macedonia, dated to start between 6600 and 6500 cal BC, it now

2 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

Fig. 1. The Early Neolithic sites in Serbia with available radiocarbon dates. a) The estimated earliest sites/dates set, b) The full data set, c) The binned data set. Site legend: 1. Biserna obala – Nosa 2. Ludoš – Budžak 3. Sajan – Domboš 4. Magareći mlin 5. Bečej 6. Ribnjak – Bečej 7. Donja Branjevina 8. Topole – Bač 9. Gospođinci 10. Sajlovo 11. Gospođinci – Futog 12. Golokut 13. Sremski Karlovci 14. Perlez – Batka 15. At – Vršac 16. Autoput E-70, P2 sever 17. Autoput Ruma – Sremska Mitrovica, km 521, deonica 4 18. Autoput E-70, km521, lokalitet 1 19. Kudoš – Šašinci 20. Baštine – Obrež 21. Vinča – Belo brdo 22. Starčevo – Grad 23. Grabovac 25. Bataševo 26. Lepenski Vir 27. Ajmana 28. Šalitrena pećina 29. Banja – Aranđelovac 30. Zmajevac 31. Rudna Glava 32. Kremenilo 33. Miokovci 35. Bakovača 36. Divostin 37. Međureč 38. Anište 39. Blagotin 40. Drenovac 41. Lazarev grad 42. Ornice 43. Selište 44. Crnoklište 45. Rudnik Kosovski 46. Svinjarička čuka 47. Pavlovac – Gumnište. seems that the earliest Neolithic sites are located further to the north as the point of origin of the expansion. (Kotsakis, 2014; Maniatis, 2014; Karamitrou-Mentessidi et al., 2015). The choice of the regression technique for this kind of analysis is not As Mavropigi is closer to the Central Balkans than Paliambela, and the straightforward and needs to be elaborated. At first glance, the solution earliest dates from both sites overlap substantially, Mavropigi is taken would be to apply the ordinary least squares (OLS) regression with

3 .Pri,e al. et Porčić, M.

Table 1 Radiocarbon sample information for the dates in the estimated earliest sites/dates set.

Site Sample description Context Lab_No C14 Age Error 68% CI 95% CI Median Mean Source

Miokovci-Crkvine Bos sp., mandibula Trench 3, BRAMS-2323 7366 29 6340–6121 6362–6098 6235 6238 Porčić et al. (2020) bottom of the pit Rudnik Kosovski Homo, cranium Grave 1 BRAMS-2413 7343 27 6242–6106 6325–6088 6197 6185 Porčić et al. (2020) Ornice-Makrešane Cervus elaphus, radius Trench 1, sq. b5; B-71 BRAMS-2223 7335 31 6237–6103 6250–6080 6174 6175 Porčić et al. (2020) Bataševo Ovis aries, scapula Trench 1, e.l. 9 BRAMS-2227 7331 27 6234–6105 6241–6089 6168 6170 Porčić et al. (2020) Zmajevac Bos taurus, calcaneus Trench 2, el. 2, pit 2 BRAMS-2264 7328 27 6233–6104 6238–6090 6166 6168 Porčić et al. (2020) Međureč Bos taurus, vertebra Trench 1, sq. 2 BRAMS-2251 7316 29 6227–6103 6232–6087 6160 6160 Porčić et al. (2020) Drenovac Mammalia, long bone Trench 15, sq. 2, BRAMS-2244 7309 28 6223–6103 6229–6084 6158 6158 Porčić et al. (2020) Next to the eastern profile Anište-Bresnica Ovis/Capra, mandibula Trench 1, pit 1 BRAMS-2331 7306 28 6221–6103 6228–6084 6157 6156 Porčić et al. (2020) Selište-Sinjac Bos taurus, phalanx III Trench 6, south-east quarter, BRAMS-2303 7300 30 6217–6105 6226–6079 6155 6154 Porčić et al. (2020) 1.3–1.5 m Bakovača-Ostra Ovis/Capra, scapula Trench 3, e.l. K BRAMS-2329 7299 27 6216–6106 6225–6081 6155 6154 Porčić et al. (2020)

4 Crnoklište Bos taurus, scapula Trench 3, northern extension, pit 9 BRAMS-2290 7293 29 6213–6106 6223–6078 6154 6152 Porčić et al. (2020) Blagotin Homo, infant bone Pit dwelling 7 OxA-8609 7270 50 6211–6075 6231–6034 6141 6139 Whittle et al. (2002) Grivac Homo, bone Trench B, layer at relative depth of 2 m Bln-869 7250 100 6219–6031 6368–5924 6130 6134 Bogdanović (1994) Sremski Karlovci, lokalitet Sonje Homo, vertebra Grave 1 BRAMS-2423 7233 28 6203–6048 6210–6027 6088 6106 Porčić et al. (2020) Marinković Lazarev grad – Crkvena građevina Cervus elaphus, humerus Beneath the wester profile, BRAMS-2225 7225 28 6102–6029 6207–6021 6077 6093 Porčić et al. (2020) 2.20 m beneath the floor level Svinjarička Čuka Charcoal Coring, 2.20 m MAMS-34883 7221 31 6101–6024 6209–6016 6089 6074 Horejs et al. (2019) Ajmana Homo, skull Burial 7 AA-58322 7219 51 6203–6019 6214–6009 6085 6099 Borić (2011) Divostin Charcoal Postohole Bln-899 7200 100 6210–5993 6343–5850 6079 6085 McPherron et al. (1988) Rudna Glava Cervus elaphus, antler tool Shaft 4f: in a channel at the northern side; 419.09 m OxA-14623 7198 36 6077–6018 6205–5998 6053 6063 Borić (2009a). Borić (2009b)

asl Journal ofArchaeologica Lepenski Vir Homo, skull Burial 122, between Buildings 47 and 47′ OxA-16005 7190 45 6082–6007 6208–5988 6051 6063 Borić (2009a). Borić (2009b) Topole-Bač Homo, costa Trench 1, skeleton 1 OxA-8693 7170 50 6071–5998 6207–5923 6039 6045 Whittle et al. (2002) Gospođinci – Nove zemlje Mammalia, long bone Emptying of the object 45 BRAMS-2367 7169 28 6055–6015 6073–5997 6035 6035 Porčić et al. (2020) Donja Branjevina Animal bone Trench 5, e.l. 6, pit GrN-15974 7155 50 6062–5992 6203–5912 6028 6029 Tasić (1993) Perlez-Batka “C” Homo, tibia Trench 2, grave 1 OxA-8605 7145 50 6061–5987 6100–5900 6020 6018 Whittle et al. (2002) Magareći mlin Animal bone Semi – pit house 1, next to the hearth Grn-15973 7130 60 6060–5925 6202–5879 6007 6003 Whittle et al. (2002) Vinogradi-Bečej Ovis/Capra, scapula Trench 2, e.l. 4, below the hearth OxA-8565 7120 55 6052–5924 6084–5884 5999 5992 Whittle et al. (2002) l Science:Reports33(2020)102528 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528 arrival time as the dependent variable (as it contains a considerable where expected values (means) of radiocarbon calibrated probability amount of error due to measurement and calibration uncertainty) and distributions are used as point estimates of the time variable. The re- distance from the origin as the independent variable. The expansion gression analyses are implemented in R (R Core Team, 2019). Specifi- speed estimate is equal to the reciprocal of the estimated slope of the cally for the RMA regression we used the lmodel2 package (Legendre, best fitting regression line. Pinhasi et al. (2005) used the OLS technique, 2018). Detailed description of the statistical analysis with the R code they provided an interval estimate of the expansion front speed with the and the spreadsheet with data used for the analysis can be found in the lower boundary defined by the slope of the distance vs. time regression, online Supplementary material 1 and Supplementary material 2 files, and the upper boundary defined by the reciprocal of the slope ofthe respectively. time vs. distance regression. The main problem for the use of the OLS is An additional complication is that the choice of the data selection that both time and distance variables contain error. The OLS regression protocol can influence the results of the regression. Ideally, wewould is applied under the assumption that only the dependent variable only use the set of the truly earliest sites, but this is not possible as the contains error, whereas the independent variable is measured without best we can get is an estimate of this set. Even though it seems intuitive error. The error in the time variable is due to uncertainty of sampling (if to use set of the estimated earliest sites/dates, this approach may po- the actual earliest date has been sampled at the site), uncertainty of the tentially bias the regression analysis. The crux of the problem is that if laboratory measurement, and the uncertainty of the calibration process. we set the terminus ante quem threshold for the selection of sites as the The error in the distance stems from the fact that we do not know the earliest, as we did here, it is more likely that the false earliest sites will true point of origin. Distances measured from an arbitrarily chosen site be selected in the south (closer to the origin of expansion) than in the are approximations in the sense that the location of a particular site is north. This is so because the time window for the inclusion of sites into taken as a proxy for the actual point of origin. For this reason, Steele the earliest group is much smaller in the north, where the threshold (2010) suggested that the Reduced Major Axis (RMA) regression value is closer to the true arrival time. The uneven spatial distribution (Legendre and Legendre, 1998: 510–517) is the most suitable regression of the false earliest sites might influence the slope of the regression line technique for the front speed expansion estimation, as it takes into (Fig. 2). account error in both variables. Estimates in Gkiasta et al. (2003) are The alternative is to use the earliest radiocarbon dates from all Early based on the RMA technique. Although Steele’s suggestion may seem Neolithic sites with available radiocarbon evidence between 6200 and like the best solution, Smith (2009) has shown that things are not that 5400 BCE (the time span of the Early Neolithic Starčevo culture) simple, as the choice of RMA over OLS cannot be justified only by the without an attempt to filter out the false earliest sites. In this case, all fact that both variables contain error. Smith (2009) identified two key other things being equal, the probability of including the false earliest issues: 1) the structure of the error (how error is distributed between sites will not change with the distance from the origin – it would be the two variables) and 2) is the relationship between the variables more or less constant throughout the study area. The slope of the re- asymmetrical (e.g. one is the cause of the other) or symmetrical (no sulting regression line would be similar to the slope of the line based clear causal relation). Smith (2009) suggests that if the error in the only on the truly earliest dates from different regions (Fig. 2). However, independent variable is small compared to the error in the dependent the intercept based on the full data set would be shifted to the more variable, the OLS is superior to RMA for the estimation of the slope. recent date in comparison with the intercept value based on the truly Likewise, if the relationship between variables is asymmetrical, again earliest dates. Counterintuitively, we should get more accurate front OLS is a better choice over RMA. This led Buchanan et al. (2011) to speed estimates by using the non-filtered data set, where the false conclude that the OLS is more appropriate than the RMA as the front earliest sites are more common, than by using the filtered data set speed estimation technique. Intuitively, it can be argued that the where the probability of including the false earliest dates is smaller but measurement error in the distance from origin is indeed smaller than it is unequally distributed in space. This kind of estimate is directly the measurement error associated with arrival times, because the point comparable to most of the earlier studies that used the same method. of origin is confined to the relatively small area, so choosing different But it comes with a price, as the use of the full data set will reduce the points within this area should not change the distance values sig- goodness of fit of the linear model due to the noise generated bythe nificantly. But it is difficult to precisely estimate the ratio of thesetwo false earliest sites in the sample. In theory, front speed estimates based errors. The issue of symmetry is also complex. We disagree with on the full set of sites should be accurate, but only if there are no other Buchanan et al. (2011) that the direction of causality is clear regarding biasing factors like the differential intensity of research and differences the distance and time – it does not seem right to argue that changes in in regional demographic histories along the gradient of expansion. arrival times are caused by changes in distance. Both distance and ar- The third option is to define an even stricter criterion for thede- rival time are changing as a result of an underlying migration process, tection of the earliest dates, which is to consider only the earliest date so they seem to be symmetrical in this respect. As this particular pro- within a certain distance bin (Hamilton and Buchanan, 2007). The blem seems to be the borderline case for the choice between RMA and downsides of this approach are that it results in a reduction of the OLS, we choose to report the results of both techniques, with arrival sample size (and increase in the uncertainty of estimates) and the loss of time as dependent and distance as the independent variable. The RMA spatial resolution as it would smooth the expansion pattern in space and estimates can be considered as the lower boundary for the speed esti- increase the fit of the linear model. The potential patterns due tonon- mate, and the OLS as the upper boundary, as RMA best fitting lines have directional or leapfrog migration would be deleted from the picture. It steeper slopes than OLS lines for the same data (we remind the reader also introduces an element of arbitrariness in choosing the bin width. that the speed is calculated as the reciprocal of the slope, therefore As all these methods have their strengths and weaknesses, the safest higher slopes translate into lower speeds and vice versa) (Smith, 2009). solution is to “triangulate” the front speed estimates by trying out all The accurate value of the front speed should be seen as being some- three approaches. Therefore, we perform regression analyses on the where in between the RMA and OLS estimates, probably closer to the three data sets (Supplementary material 2): OLS estimate, given the assumption that the error in arrival time is higher than error in distance. 1. The estimated earliest sites/dates set. This is the same set of dates Calibrated radiocarbon dates are not point estimates but probability and sites used to answer the first two questions in this study (de- distributions, therefore we performed a series of regressions with dif- scribed above). ferent possible point estimates of calendar dates following Steele’s 2. The full data set. The set of earliest radiocarbon dates from each (2010) and Hamilton and Buchanan’s (2007) Monte Carlo resampling Early Neolithic site that were radiocarbon dated within the study approach. In addition to performing series of regressions based on re- area. This sample consists of 47 dates from 47 sites. 26 out of 47 are sampling, we also performed and reported the results of regressions the new dates generated by the BIRTH project (Supplementary

5 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

dates set.

3. Results

The calibration of the radiocarbon dates from the estimated earliest sites set is presented in Fig. 3. The oldest date comes from the site of Miokovci-Crkvine in Western Serbia (6362–6098 cal BC at 95%, expected value 6238 BCE), but it is statistically indistinguishable (p = 0.572, based on the chi-square test performed using the Combine function in OxCal) from the second oldest date from Rudnik Kosovski (6325–6088 cal BC at 95%, expected value 6185 cal BC) in Kosovo, which is the southernmost early site in our sample. These two dates are followed by a cluster of dates from the Central and Southern Serbia with the 95% confidence intervals between ~6200 and ~6000 cal BC. Ajmana and Lepenski Vir from the Danube Gorges region (Eastern Serbia) also fall into this interval, although means of their probability distributions are closer to 6100 cal BC. Most of the latest dates (with expected values of ~6000 cal BC) all come from the north. The spa- tiotemporal pattern of the expansion is clearly revealed in the map based on the expected values of the earliest radiocarbon dates for each grid cell (Fig. 4). There is a general south-north gradient suggesting that the expansion followed the expected route, along the South Morava and Great Morava river valleys. However, this gradient is not perfect, it is a statistical trend, as there are sites to the north with earlier dates than sites to the south. The mean front speed estimates of the RMA regression analyses (the summary of all regression analyses is given in Table 2) based on 10,000 resampled calendar date configurations for the estimated earliest sites/ dates, the full and the binned data sets are 1.22 km/year, 0.43 km/year, and 1.43 km/year, respectively (full distributions in Fig. 5). The mean front speed estimates of the 10,000 OLS regressions based on the resampled calendar date configurations for the estimated earliest sites/ dates, the full and the binned data sets are 2.18 km/year, 0.97 km/year, and 2.44 km/year, respectively (Fig. 5). The mean goodness of fit of the linear model is measured by the mean value of Pearson's correlation coefficient across different realizations of the resampled calendar dates. It is the highest for the binned data set (r = 0.67), followed by the earliest sites set (r = 0.55), and it is the lowest for the full data set (r = 0.44). If we only use the means of calibrated probability distributions of individual radiocarbon dates as point estimates of the Neolithic arrival time, the results of the RMA regression are as follows (Table 3, Fig. 6). For the estimated earliest sites/dates set, the estimated front speed is 1.57 km/year, for the full data set it is 0.39 km/year, and for the binned data set it is 1.64 km/year. For the OLS regressions based on point estimates of arrival time the front speeds are 2.24 km/year, 0.95 km/ year, and 2.04 km/year, for the same three data sets, respectively. The Fig. 2. Schematic representation of the effects of different kinds of data selec- linear correlation coefficient is the highest for the binned data set tion on the estimated regression curve in time vs. distance analysis: a) The earliest radiocarbon dates are included from all sites, b) The earliest radio- (r = 0.81); it is slightly lower for the estimated earliest sites/dates set carbon dates included only from the sites with the earliest dates older than the (r = 0.7); and the lowest value is associated with the full data set defined threshold. (r = 0.41). For all three data sets with point estimates the regression models are statistically significant at the 0.05 level. material 2). 4. Discussion 3. The binned data set consists of the 8 earliest radiocarbon dates from 8 distance bins (6 are the new BIRTH dates). This is based on the Before evaluating the time of the arrival of the Neolithic to the method from Hamilton and Buchanan (2007). We first grouped the Central Balkans, a note needs to be made about estimating the earliest sites from the estimated earliest sites/dates set into 50 km distance occupation in a region or on a site. The probability of actually sampling bins (Supplementary material 2). These bins should be imagined as the earliest date from a site will depend on the population dynamics concentric 50 km wide annuli emanating from the assumed origin of and duration of a particular settlement as well as on the sample size expansion. We then selected the earliest date from each bin as the (Perreault, 2011). For example, if we assume, in accord with the the- arrival time of the Neolithic population to the particular spatial area oretical and empirical results regarding the Neolithic Demographic defined by the bin. Transition, that the Neolithic population size was increasing through time (Bocquet-Appel and Bar-Yosef, 2008; Shennan, 2018), then the Therefore, the estimated earliest sites/dates set is a subset of the full most probable date to sample will be a date from the period when the data set, and the binned set is a subset of the estimated earliest sites/ population size was at its peak. The implication is that it is highly

6 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

Fig. 3. The calibrated radiocarbon dates from the estimated earliest sites/dates set in the Central Balkans. unlikely to sample the earliest dates from any site. It is difficult to relatively short time, in circumstances related to the 8.2 ky event precisely calculate the expected offset without assuming a population (Weninger et al., 2006; Berger and Guilaine, 2009; Pross et al., 2009; dynamics model and knowing the duration of each settlement, but a Krauß et al., 2018). This would result in the observed pattern where the conservative educated guess would be that the actual earliest dates are earliest dates from Serbia and North Macedonia are almost the same, usually few decades earlier than the sampled ones. Having this fact in but until more dates from the North Macedonia become available, it is mind, we tentatively conclude that it is most probable that the Neolithic more parsimonious to explain this pattern as a result of an extremely arrived at the Central Balkans between 6250 and 6200 cal BC. small number of the Early Neolithic dates (N = 29) from a small It is interesting to note that the oldest dates from Serbia are older number of sites (N = 8) in North Macedonia (for a full list of published than most of the radiocarbon dates for the Early Neolithic in the North Macedonian dates see Fidanoski, 2009; Naumov, 2009). In order Republic of North Macedonia, which is immediately to the south of to resolve this issue, new research programmes (e.g. such as Horejs Serbia. There are pre-6250 cal BC radiocarbon dates coming from the et al., 2019) need to be funded in order to fill the large gap in the North Macedonian sites Amzabegovo and Čuka-Topolčani (Naumov, quantity and quality of archaeological information related to the spread 2009; Fidanoski, 2009), but the accuracy and precision of these dates of the Neolithic between Northern Greece and Central Serbia. are problematic as these are conventional radiocarbon dates made on As for the front speed estimates, first we will discuss the large dis- charcoal with very large standard errors (over 150 radiocarbon years). crepancy in the estimated front speeds between the full data set on one It is theoretically possible that the Central Balkans was populated by a side, and the estimated earliest and binned sets on the other (Table 2). It long-distance migration from northern Greece to southern or even is absolutely certain that the average speed of expansion must have central Serbia ~6250 cal BC, “jumping over” the territory of North been higher than the RMA estimates of ~0.4 km/year, because at such Macedonia or sweeping across North Macedonia and Serbia in a rate, it would take more than thousand years for the Neolithic to cross

7 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

Fig. 4. The spatio-temporal pattern of the expansion based on the estimated earliest sites/dates set. The color of each grid cell depends on the median of the earliest Neolithic date in the cell.

Table 2 Summary of the front speed estimates based on the Monte Carlo resampling of the arrival time calendar dates for the three data sets.

RMA mean front speed RMA mean intercept value OLS mean front speed OLS mean intercept value Mean Pearson's r estimate (km/year) (years cal BC) estimate (km/year) (years cal BC)

The estimated earliest sites/ 1.22 6473 2.18 6310 0.55 dates set 95% CI 95% CI 95% CI 0.97–1.54 1.53–3.78 0.34–0.73 The full set 0.43 7025 0.97 6412 0.44 95% CI 95% CI 95% CI 0.4–0.45 0.85–1.11 0.39–0.49 The binned set 1.43 6456 2.44 6353 0.67 95% CI 95% CI 95% CI 0.97–2.21 1.25–5.76 0.3–0.92

8 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

Fig. 5. The distribution of front speed estimates coefficients based on 10,000 RMA and OLS regressions with each fitted to a different set of Ndrawssingle calendar-year values, where N is the number of radiocarbon dates from a specific data set.

9 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

Table 3 Summary of the front speed estimates based on the point estimates of the arrival time calendar dates for the three data sets.

RMA front speed estimate RMA intercept estimate (years OLS front speed estimate OLS intercept estimate (years Pearson's r (km/year) cal BC) (km/year) cal BC)

The estimated earliest sites/dates 1.57 6387 2.24 6304 0.7 set 95% CI 95% CI 1.17–2.11 1.57–3.9 The full set 0.39 7113 0.95 6407 0.41 95% CI 95% CI 0.3–0.51 0.57–2.85 The binned set 1.64 6404 2.04 6352 0.81 95% CI 95% CI 0.94–2.88 1.18–7.6

~450 km from the southernmost to the northernmost part of the study 2004), therefore different combinations of these two parameters can area, which cannot possibly be true. The OLS estimates based on the full result in different speed values. This means that the front expansion set are at first glance realistic, as they are consistent with the 1km/year speeds around 1.5–2 km/year are possible within the limits of the prediction of the standard WoA model and estimates from previous parameter values of the WoA model that are considered to be possible studies (e.g. Bocquet-Appel et al., 2012; Gkiasta et al., 2003; Pinhasi and realistic (Ammerman and Cavalli-Sforza, 1984: 81). et al., 2005), but on closer analysis the OLS estimates seem to be un- The more recent formulations of the WoA model that include cul- derestimates, as well. The only way to get a value that is comparably tural diffusion suggest that the front speed is also proportional tothe low as the OLS estimate is to assume that the Neolithic arrived to the degree of cultural diffusion (Fort, 2012). Therefore, relatively high es- southernmost point of our study area as early as 6362 cal BC (the upper timates of the front speed in the Central Balkans could have arisen from limit of the 95% CI of the oldest date in the earliest sites/dates set), and the increased conversion of the local hunter-gatherers to farmers as a that it reached the northernmost point as late as 5879 cal BC (the lower result of the cultural diffusion, but the problem with this interpretation limit of the 95% CI of the oldest date in the earliest sites/dates set). In is that there is no secure evidence of the Mesolithic presence in the this case the average speed would be ~1.1 km year, which is still higher Central Balkans outside the microregion of the Danube Gorges (Gurova than the OLS estimate. But this scenario is highly unlikely and borders and Bonsall, 2014). on the impossible, as the lower boundary of the 95% confidence in- With these results we can only make general remarks about the tervals for the oldest date from a site which is situated ~130 km to the mode of expansion, i.e. the underlying demographic, economic and north of the study area is 5883 cal BC (see Supplementary material 1). social processes driving the expansion. The moderately good fit of the This leads us to conclude that the full set of sites is biased. Fig. 6 sug- linear model suggests that the WoA is a satisfactory general model of gests that there are far more sites in the north than in the south, the expansion in the Central Balkans. At the lower spatio-temporal especially sites with later founding dates. This is tilting the regression scales the process seems to carry significant noise which reduces the line towards lower speeds. This difference in the number of sites be- goodness of fit of the model. This can be interpreted in two mutually tween the south and the north is most probably a consequence of the non-exclusive ways. The first possibility is that the earliest farming differential intensity of research (the number of investigated Early communities sometimes practiced relatively long-distance migrations Neolithic sites with radiocarbon dates is lower in the south than in the (e.g. ~100 km opposed to the maximum of ~50 km assumed in the north, see Fig. 1) and/or potentially different regional demographic standard WoA model) corresponding to the leapfrog colonization model histories. Therefore, front speed estimates based on the full set of sites (van Andel and Runnels, 1995; Zvelebil, 2001). The second possibility can safely be rejected as unrealistically low. We proceed with the dis- is that these discrepancies reflect the sampling effects and differential cussion of the front speed estimates based only on the estimated earliest regional intensity of research. In conjunction with the fact that the sites and binned data sets. earliest dates from sites are unlikely to be sampled a priori, this might The front speed estimates based on the Monte Carlo resampling are have blurred the south-north gradient. As already mentioned, the in- either broadly consistent (RMA estimate on the earliest sites/dates set) clusion of the false earliest sites would also decrease the goodness of fit, with, slightly higher (RMA estimate on the binned data set) than, or which is apparent here with the full data set where the correlation significantly above (OLS estimates) the continental front speed estimate between the time of arrival and distance is the lowest in this study. ranges from previous studies (e.g. Gkiasta et al., 2003; Pinhasi et al., Under certain conditions, intercept values of the best-fitting re- 2005; Bocquet-Appel et al., 2012). However, the estimates from the gression line can be used to evaluate the accuracy of the model. The previous studies were based on the point estimates of arrival times. intercept value can be interpreted as the estimated date for the start of Strictly speaking, we should only compare them to the theoretical ex- the expansion which can be compared to the actual dates associated pectations for the WoA model. In this case we can conclude that they with the assumed point of origin. However, this would require us to are considerably higher than the often-cited value of ~1 km/year make some very strong assumptions that would make the entire effort predicted by the standard parametrization of the WoA model speculative, therefore we decided not to follow this path at this moment (Ammerman and Cavalli-Sforza: 80, 1984; Hazelwood and Steele, (see Supplementary material 1 for a detailed explanation). 2004). On the other hand, point estimates are comparable to the empirical estimates from other studies, and they unequivocally suggest that the Neolithic front expansion speed in the Balkans was around 5. Conclusion ~1.5–2 times higher than the continental average of ~1 km/year, although not as high as suggested by Biagi et al. (2005). The first farmers arrived at the Central Balkans as earlyas The fact that the estimated speed is higher than the standard 1 km/ 6250–6200 cal BC and reached the Great Pannonian Plain by 6000 cal year does not mean that is not consistent with the WoA model. The BC. The spread of the Neolithic across this region generally unfolded in speed of the expansion is proportional to the square root of the product the south-north direction, following the major river courses. Our results between the migratory activity and the intrinsic growth rate indicate that, in general, the standard WoA model of expansion is ac- (Ammerman and Cavalli-Sforza: 68, 1984; Hazelwood and Steele, ceptable as the first approximation of the process of expansion, with the possibility of sporadic leapfrog migration events. The Neolithic

10 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

expansion in the Central Balkans was faster compared to the continental average.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This study is a result of the project BIRTH: Births, mothers and babies: prehistoric fertility in the Balkans between 10,000 and 5000 BC funded by the European Research Council within the European Union's Horizon 2020 research and innovation programme (Grant Agreement No. 640557, Principal Investigator: Sofija Stefanović). We are most grateful to the anonymous reviewers and to the editor Dr. Andy Howard for their constructive comments and suggestions. We would also like to thank Prof. Dr. Vesna Dimitrijević, Dr. Ivana Živaljević, and Dr. Jelena Jovanović for their expertise regarding the collection and bioarchaeo- logical analysis of animal and human bones used as samples for radiocarbon dating within the BIRTH project. The responsibility for any remaining omissions and errors is exclusively ours.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https:// doi.org/10.1016/j.jasrep.2020.102528.

References

Ammerman, A.J., Cavalli-Sforza, L.L., 1971. Measuring the rate of spread of early farming in Europe. Man 6, 674–688. Ammerman, A.J., Cavalli-Sforza, L.L., 1973. A population model for the diffusion of early farming in Europe. In: Renfrew, C. (Ed.), The Explanation of Culture Change: Models in Prehistory. Duckworth, London, pp. 343–357. Ammerman, A.J., Cavalli-Sforza, L.L., 1984. The Neolithic Transition and the Genetics of Populations in Europe. Princeton University Press. Anders, A., Siklósi, Z. (Eds.), 2012. The First Neolithic Sites in Central/South-East European Transect. Volume III: The Körös Culture in Eastern Hungary. British Archaeological Reports, Oxford, UK. Berger, J.-F., Guilaine, J., 2009. The 8200 calBP abrupt environmental change and the Neolithic transition: a Mediterranean perspective. Quat. Int. 200, 31–49. Biagi, P., Shennan, S., Spataro, M., 2005. Rapid rivers and slow seas? New data for the radiocarbon chronology of the Balkan Peninsula. In: Nikolova, L., Higgins, J. (Eds.), Prehistoric Archaeology and Anthropological Theory and Education. International Institute of Anthropology, Salt Lake City, pp. 41–51. Bocquet-Appel, J.-P., Bar-Yosef, O. (Eds.), 2008. The Neolithic Demographic Transition and its Consequences. Springer, Berlin. Bocquet-Appel, J.-P., Naji, S., Vander Linden, M., Kozlowski, J., 2012. Understanding the rates of expansion of the farming system in Europe. J. Archaeol. Sci. 39, 531–546. Bogdanović, M., 1994. Prilog proučavanju apsolutne hronologije protostarčevačke i starčevačke kulture. Starinar 47, 187–192. Borić, D., 2009a. Absolute dating of metallurgical innovations in the Vinča culture of the Balkans. In: Kienlin, T., Roberts, B. (Eds.), Metals and Societies: Studies in Honour of Barbara S Ottaway. Dr Rudolf Habelt GMBH, Bonn, pp. 191–245. Borić, D., 2011. Adaptations and transformations of the Danube Gorges Foragers (c. 13.000-5500 BC): an overview. In: Beginnings – New Research in the Appearance of the Neolithic between Northwest Anatolia and the Carpathian Basin. Papers of the International Workshop 8-9.4.2009. Istanbul, pp. 157–203. Borić, D., Dimitrijević, V., 2009b. Apsolutna hronologija i stratigrafija Lepenskog Vira. Starinar LVII/2007, 9–55. Borić, D., Price, T.D., 2013. Strontium isotopes document greater human mobility at the start of the Balkan Neolithic. Proc. Natl. Acad. Sci. 110, 3298–3303. Brami, M., Zanotti, A., 2015. Modelling the initial expansion of the Neolithic out of Anatolia. Documenta Praehistorica 42, 103–116. Bronk Ramsey, C., 2009. Bayesian analysis of radiocarbon dates. Radiocarbon 51, 337–360. Buchanan, B., Hamilton, M., Edinborough, K., O’Brien, M.J., Collard, M., 2010. A com- ment on Steele’s (2010) “Radiocarbon dates as data: quantitative strategies for estimating colonization front speeds and event densities. J. Archaeol. Sci. 38 (9), Fig. 6. RMA and OLS regression of means of calibrated dates, as point estimates 2116–2122. of the time of arrival vs. spatial distances from the assumed origin of the ex- Dolukhanov, P., Shukurov, A., Gronenborn, D., Sokoloff, D., Timofeev, V., Zaitseva, G., pansion at Mavropigi. Red line – fitted regression line; gray lines – 95% CI limits 2005. The chronology of Neolithic dispersal in Central and Eastern Europe. J. for the regression line. Archaeol. Sci. 32, 1441–1458. Douka, K., Efstratiou, N., Hald, M.M., Henriksen, P.S., Karetsou, A., 2017. Dating Knossos

11 M. Porčić, et al. Journal of Archaeological Science: Reports 33 (2020) 102528

and the arrival of the earliest Neolithic in the southern Aegean. Antiquity 91, Central Serbia. University of Pittsburgh, Pittsburgh, pp. 379–387. 304–321. Naumov, G., 2009. The process of Neolithization. In: Naumov, G., Fidanovski, L., Fidanoski, L., 2009. Periodization and chronology of the Neolithic cultures. In: Naumov, Tolevski, I., Ivkovska, A. (Eds.), Neolithic communities in the Republic of Macedonia. G., Fidanovski, L., Tolevski, I., Ivkovska, A. (Eds.), Neolithic Communities in the Dante, Skopje, pp. 17–27. Republic of Macedonia. Dante, Skopje, pp. 31–34. Perlès, C., 2001. The Early Neolithic in Greece: The First farming Communities in Europe. Fort, J., 2012. Synthesis between demic and cultural diffusion in the Neolithic transition Cambridge University Press, Cambridge. in Europe. Proc. Natl. Acad. Sci. 109, 18669–18673. Perlès, C., Quiles, A., Valladas, H., 2013. Early seventh-millennium AMS dates from do- Fort, J., Pujol, T., Vander Linden, M., 2012. Modelling the Neolithic Transition in the mestic seeds in the Initial Neolithic at Franchthi Cave (Argolid, Greece). Antiquity 87, Near East and Europe. Am. Antiq. 77, 203–219. 1001–1015. Garašanin, M., 1982. The stone age in the Central Balkan Area. Part 1 In: Boardman, J., Perreault, C., 2011. The impact of site sample size on the reconstruction of culture his- Edwards, I.E.S., Hammond, N.G.L., Sollberger, E. (Eds.), Cambridge Ancient History. tories. Am. Antiq. 76, 547–572. Cambridge University Press, Cambridge, pp. 75–135. Pinhasi, R., Fort, J., Ammerman, A.J., 2005. Tracing the origin and spread of agriculture Gatsov, I., Boyadzhiev, Y. (Eds.), 2009. The First Neolithic Sites in Central/South-East in Europe. PLoS ONE 3, e2220–e2228. European Transect, Volume I. Early Neolithic Sites on the Territory of Bulgaria. Porčić, M., Blagojević, T., Pendić, J., Stefanović, S., 2020. The Neolithic Demographic Archaeopress, Oxford. Transition in the Central Balkans: population dynamics reconstruction based on new Gkiasta, M., Russell, T., Shennan, S., Steele, J., 2003. Neolithic transition in Europe: the radiocarbon evidence. Philos. Trans. R. Soc. B: Biol. Sci. https://doi.org/10.1098/ radiocarbon record revisited. Antiquity 77, 45–62. rstb.2019.0712. Gurova, M., Bonsall, C., 2014. ‘Pre-Neolithic’ in Southeast Europe: a Bulgarian perspec- Pross, J., Kotthoff, U., Müller, U., Peyron, O., Dormoy, I., Schmiedl, G., Kalaitzidis, S., tive. Documenta Praehistorica 41, 95–109. Smith, A., 2009. Massive perturbation in terrestrial ecosystems of the Eastern Hamilton, M.J., Buchanan, B., 2007. Spatial gradients in Clovis-age radiocarbon dates Mediterranean region associated with the 8.2 kyr BP climatic event. Geology 37, across North America suggest rapid colonization from the north. Proc. Natl. Acad. Sci. 887–890. 104 (40), 15625–15630. R Core Team, 2019. R: A language and environment for statistical computing. R Hazelwood, L., Steele, J., 2004. Spatial dynamics of human dispersals: constraints on Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project. modelling and archaeological validation. J. Archaeol. Sci. 31 (6), 669–679. org/. Henderson, D.A., Baggaley, A.W., Shukurov, A., Boys, R.J., Sarson, G.R., Golightly, A., Reingruber, A., Toufexis, G., Kyparissi-Apostolika, N., Anetakis, M., Maniatis, Y., 2014. Regional variations in the European Neolithic dispersal: the role of the coast- Facorellis, Y., 2017. Neolithic Thessaly: radiocarbon dated periods and phases. lines. Antiquity 88, 1291–1302. Documenta Praehistorica 44, 34–53. Hofmanová, Z., et al., 2016. Early farmers from across Europe directly descended from Robb, J., Miracle, P., 2007. Beyond 'migration' versus 'acculturation': new models for the Neolithic Aegeans. Proc. Natl. Acad. Sci. 113, 6886–6891. spread of agriculture. In: Whittle, A., Cummings, V. (Eds.), Going Over: The Horejs, B., Bulatović, A., Bulatović, J., Brandl, M., Burke, C., Filipović, D., Milić, B., 2019. Mesolithic-Neolithic Transition in North-West Europe. Oxford University Press, New Insights into the Later Stage of the Neolithisation Process of the Central Balkans. Oxford, pp. 99–115. First Excavations at Svinjarička Čuka 2018. Archaeol. Austriaca 103, 175–226. Shennan, S., 2018. The First Farmers of Europe: An Evolutionary Perspective. Cambridge Karamitrou-Mentessidi, G., Efstratiou, N., Kaczanowska, M., Kozłowski, J.K., 2015. Early University Press, Cambridge. neolithic settlement of Mavropigi in Western Greek Macedonia. Eurasian Prehistory Smith, R.J., 2009. Use and misuse of the reduced major axis for line-fitting. Am. J. Phys. 12, 47–116. Anthropol. 140, 476–486. Kotsakis, K., 2014. Domesticating the periphery. Pharos 20, 41–73. Steele, J., 2010. Radiocarbon dates as data: quantitative strategies for estimating colo- Krauß, R., Marinova, E., De Brue, H., Weninger, B., 2018. The rapid spread of early nization front speeds and event densities. J. Archaeol. Sci. 37, 2017–2030. farming from the Aegean into the Balkans via the Sub-Mediterranean-Aegean Tasić, N.N., 1993. Nekoliko novih radiokarbonskih datuma. Glasnik SAD 9, 99–102. Vegetation Zone. Quat. Int. 496, 24–41. van Andel, T.H., Runnels, C.N., 1995. The earliest farmers in Europe. Antiquity 69, Legendre, P., Legendre, L.F.J., 1998. Numerical Ecology. Elsevier, Amsterdam. 481–500. Legendre, P. 2018. lmodel2: Model II Regression. R package version 1.7-3. < https:// Weninger, B., Alram-Stern, E., Bauer, E., Clare, L., Danzeglocke, U., Jöris, O., Kubatzki, CRAN.R-project.org/package=lmodel2 > . C., Rollefson, G., Todorova, H., van Andel, T., 2006. Climate forcing due to the 8200 Luca, S.A., Suciu, C. (Eds.), 2011. Early Neolithic (Starčevo-Criş) Sites on the Territory of cal yr BP event observed at Early Neolithic sites in the eastern Mediterranean. Quat. Romania. Archaeopress, Oxford. Res. 66, 401–420. Maniatis, Y., 2014. Radiocarbon dating of the major cultural changes in Prehistoric Whittle, A., Bartosiewicz, L., Borić, D., Pettitt, P., Richards, M., 2002. In the beginning: Macedonia: recent developments. 1912–2012. A Century of Research in Prehistoric new radiocarbon dates for the Early Neolithic in northern Serbia and south-east Macedonia. In: Stefani, E., Merousis, N., Dimoula, A. (Eds.), Proceedings of the Hungary. Antaeus 63–117. International Conference, Archaeological Museum of Thessaloniki. Archaeological Zilhão, J., 2003. The Neolithic transition in Portugal and the role of demic diffusion in the Museum of Thessaloniki, Thessaloniki, pp. 22–24. spread of agriculture across West Mediterranean Europe. In: Ammerman, A.J., Biagi, Mathieson, I., et al., 2018. The genomic history of southeastern Europe. Nature 555, P. (Eds.), The Widening Harvest: The Neolithic Transition in Europe. Archaeological 197–203. Institute of America, Boston, pp. 207–223. McPherron, A., Bucha, V., Aitken, M.J., 1988. Absolute dating of Divostin, Grivac-Barice Zvelebil, M., 2001. The agricultural transition and the origins of Neolithic society in and Banja. In: McPherron, A., Srejović, D. (Eds.), Divostin and the Neolithic of Europe. Documenta Praehistorica 28, 1–26.